box_tree: make NodeId tuple fields private, use named methods to access

[tomcur]

No description provided.

[waywardmonkeys]

Would this be better as a struct-struct with 2 fields rather than a tuple struct?

[tomcur]

Would this be better as a struct-struct with 2 fields rather than a tuple struct?

Yes, that makes sense. There may also be value in thinking about the memory layout a bit, to allow methods like id_is_newer to be compiled down to a single u64 comparison without requiring any shifting of memory (this type is a likely target for exposing through FFI in the future, I think, so it'll get repr(C) anyway). In Vello sparse we do some cfg based on endianness, but potentially here we can decide to just do the optimization assuming the target is little endian.

[tomcur]

I've given the struct named fields, but haven't made them pub (yet?) for two reasons: first, the idx() method casts to usize, which may be nice to keep for ergonomics, regardless of the exact internal representation. Second, we may still want to have a deterministic layout / packing of this struct, but perhaps we need to think more about the "ID is newer" semantics: that's currently based on generation first, which is perhaps fine for just having a deterministic ordering, but generations are only meaningfully comparable if the node has the same idx (as generation tends to be tracked per-idx).

[waywardmonkeys]

Overall, this is the right thing ... but feel like maybe we need to think about some of these details a little.